Glossary · Term

counterfactual regret minimization

← all terms

Definition

Plain language

The self-play method behind superhuman poker bots, where the program plays itself millions of times and drifts toward the choices it least regrets.

As stated in the literature

CFR — an iterative algorithm that minimizes per-decision regret across self-play to converge on a Nash-equilibrium strategy in imperfect-information games; the basis of solver-built poker AIs like Libratus, contrasted with PokerSkill's training-free approach.

Also called: CFR

Why it matters: It's the engine that produced superhuman poker play, showing how self-play alone can converge on a strategy no opponent can beat.

For example, a poker program plays itself millions of times and gradually leans toward the bets it regretted not making, until it stops being exploitable.

Heard on the show

“The famous superhuman bots — Libratus is the headline one — were built on an algorithm called counterfactual regret minimization.”
Episode 100 — How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert

Mentioned in 1 episode

  1. 100
    How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert

Related terms